Wall temperature prediction at critical heat flux using a machine learning model

To determine heat transfer regimes of the pre and post CHF, the SPACE code calculates the wall temperature from a nucleate boiling heat transfer model at the given CHF. It needs iterations and consumes a large amount of computing time. To reduce the calculation time, this paper introduces the application of a machine learning method. Big data of the wall temperature at CHF was built by using the subprogram constructed as is in the SPACE code. Based on that database, the neural network models were trained and two neural network models having different configurations were suggested. The developed neural network models were implemented in the SPACE code and test calculations were performed. The neural network applied SPACE code properly predicted the wall temperature at CHF. In test calculations, the calculation time was also investigated. All suggested neural network models highly enhanced the calculation speed corresponding to a maximum 86% time reduction. (C) 2020 Elsevier Ltd. All rights reserved.